A Harvard study shows AI model can outperform physicians in emergency room diagnoses

A Harvard-led study found an OpenAI reasoning model could diagnose some emergency cases more accurately than physicians when given the same limited records.

The tests used real emergency and benchmark clinical cases, suggesting strong performance in high-pressure, incomplete-information settings.

Researchers noted the system worked only from text records, so its real-world value in live care is still unproven.

Before any deployment, the model needs prospective clinical trials to confirm whether it improves patient outcomes.